NVIDIA links BioNeMo to Anthropic’s Claude Science workbench to speed life-science AI workflows

NVIDIA said it is integrating its BioNeMo Agent Toolkit into Anthropic’s Claude Science, a science-focused AI workbench that lets researchers use natural language to direct agent-driven workflows. NVIDIA’s aim, the company said, is to make GPU-accelerated life-science capabilities accessible inside the same environment where researchers are planning and iterating on experiments, rather than forcing them to stitch together models, endpoints, and software configuration themselves.

Announced in a blog post dated June 30, NVIDIA said Claude Science “converses” with agents in natural language to help scientists run work end to end. Within that workflow, NVIDIA’s BioNeMo Agent Toolkit is positioned as a resource Claude Science can access, packaged as callable “skills” that Claude can select, prepare inputs for, and execute as part of a broader scientific task.

NVIDIA described the BioNeMo Agent Toolkit as “accelerated capabilities” connected to NVIDIA compute resources deployed anywhere. The toolkit, according to NVIDIA, is meant to bring NVIDIA accelerated models, libraries, and NVIDIA NIM microservices into Claude Science’s agent environment, so that the computationally heavy steps within a research pipeline run faster while other parts of the workflow remain oriented around scientific reasoning and review.

The company said that life sciences is entering an era of “computational scale” and that NVIDIA has been building a GPU-accelerated stack for more than a decade across hardware, frameworks, libraries, models, microservices, and domain tools. In that context, NVIDIA framed the integration as an “iterative loop” in which scientists can inspect outputs, refine questions, and determine the next step while the AI system continues orchestrating and running the computational parts.

NVIDIA cited adoption by major pharmaceutical firms, saying “18 of the top 20” pharmaceutical companies use NVIDIA BioNeMo. It also pointed to use cases spanning drug discovery, genomics, medical imaging, molecular design, and protein engineering, arguing that the underlying ecosystem already supports multiple stages of modern life-science AI research.

For the Claude Science integration itself, NVIDIA said scientists begin by describing a task such as analyzing a genomic sequence, predicting a protein structure, or designing a potential binder. Claude Science then interprets the request and routes it through preconfigured, domain-specialized agents that know established workflows across areas including genomics, proteomics, single-cell analysis, cheminformatics, and clinical research.

NVIDIA said the BioNeMo Agent Toolkit provides the agents with context to connect each workflow step to an appropriate NVIDIA scientific capability. Each toolkit skill includes details about its purpose and required inputs, which NVIDIA says helps agents prepare valid inputs and execute the workflow, returning outputs for researcher review. NVIDIA described an example focused on inhibitor discovery for common cancer targets, where a scientist provides a known cancer-causing antigen mutation and asks Claude to design numerous potential inhibitors, with Claude Science and BioNeMo accelerating high-throughput prediction, optimization, and validation through NVIDIA NIM microservices.

Claude Science is entering public beta, NVIDIA said, with Anthropic inviting researchers to provide feedback on additional domain specialists and integrations they want to see. NVIDIA added that BioNeMo Agent Toolkit is “open and harness-agnostic,” meaning, in the company’s framing, the same scientific skills can work across agent frameworks and research platforms.

For all the operational detail in NVIDIA’s description, several specifics were not included in the blog post. NVIDIA did not lay out performance benchmarks, availability terms beyond developer resources and GitHub, pricing, or which exact model versions will be surfaced first inside the beta. The post also did not clarify how users will handle data governance, permissions, or auditability for sensitive biomedical workflows once agents start selecting and executing accelerated steps automatically.

Looking ahead, researchers and developers using Claude Science’s public beta will likely watch for which BioNeMo-powered domain skills are offered first, how reliably the agents choose the correct accelerated workflow steps, and whether NVIDIA’s toolkits broaden to additional life-science specializations over time. For the companies involved, the integration also tests a broader strategy in the AI industry: moving from stand-alone models to agent-driven systems that can plan with tools, then execute compute-intensive steps quickly and iteratively within the same research workflow.